Photo Credit: Google MusicFX
Google reports on the successes of its generative AI tools MusicFX and TextFX, and announces the release of ImageFX. More than 10 million tracks have already been created using Google’s MusicFX.
As artificial intelligence tools ripple transformative changes throughout creative industries worldwide, Google is investing an increasing amount of time and funds into its own GenAI tools. Its suite of new generative tools, MusicFX, TextFX, and ImageFX, released in limited capacities last year, are now available for general use — and their output is getting better, albeit indicative of their respective training sets.
MusicFX, the upgraded version of Google’s music-generating tool, MusicLM, released last year to a limited audience, can generate tracks or loops of up to 70 seconds in length. Since its launch, the tool has been used to create over 10 million tracks, and Google has implemented changes based on user feedback, with “higher quality” audio and faster output than its previous iteration.
Like MusicLM before it, MusicFX lets users enter a text prompt to describe the song they want to generate. The tool then creates two 30-second versions, with options to increase the length to 50 or 70 seconds or stitch the beginning and end together to create a seamless loop. MusicFX provides a word cloud of additional words for relevant descriptions to enable users to further tweak the output with various tempos and instruments.
The output is about what you’ve probably come to expect from genAI tools of this nature. TechCrunch reports that “music generation tools are getting to the point where it’s tough for this writer to distinguish between the outputs.” So the tracks produced are “impressively clean, crisp-sounding” — but lack the heart and soul of human-created music, and often end up sounding “boring, uninspired, and melodically unfocused.”
Songs generated with stringed instruments sound noticeably worse, “like a cheap MIDI sample,” which is probably indicative of MusicFX’s limited training set. DJ-inspired tracks come across much better, but still seem to be lacking that “je ne sais quois” that makes listening to music so meaningful and personal.
MusicFX uses SynthID, an inaudible watermarking technology developed by Google’s Deepmind division, which makes it clear which tracks were generated using the tool. To avoid violating copyrights, the tool filters prompts that mention specific artists or vocals.
Google asserts that the launch of its genAI tools extends the company’s commitment to developing technologies responsibly and in-line with Google’s AI Principles. The tools are available in Google’s AI Test Kitchen, the company’s app that allows users to try out experimental AI-based systems, currently enabled for users in the US, Australia, New Zealand, and Kenya.
